What AI Can Actually Do for a Solo Law Practice

Start with the jobs, not the hype

The fastest way to be disappointed by AI is to ask it to be a lawyer. The fastest way to get value is to point it at the bounded, repetitive tasks that fill a solo's day and don't actually need judgment — just speed and a careful second read.

So this guide is organized around concrete jobs, each tied to a real solo workflow, with the limits stated as plainly as the capabilities. Treat the capabilities as "things that save you time when supervised," not "things you can hand off and forget." That distinction is the whole game.

Capability 1: Reviewing a contract or document

This is the anchor capability for most solo practices, because contracts are where the hours go.

The workflow. A small-business client forwards a 30-page vendor agreement and asks "is this okay to sign?" The traditional path is reading it cover to cover, flagging concerns as you go. With AI, you start by having it review the document and surface what's notable: unusual terms, one-sided provisions, clauses that are conspicuously missing, language that doesn't match what the client described. You get a structured first pass in seconds, with pointers to the exact sections.

What that buys you. Not a conclusion — a map. Instead of reading thirty pages cold, you read them knowing where the friction is, and you spend your attention on the three provisions that actually matter for this client. The review still happens; it just starts oriented instead of blank.

The boundary. The tool flags that an indemnification clause is unusually broad. Whether that's a problem for this client, in this deal is your call, not the model's. It surfaces; you decide.

Capability 2: Extracting clauses and details

Closely related, but worth separating, because extraction is where AI is most reliably strong.

The workflow. You need every defined term and where it's used, all the dates and deadlines, the renewal and termination mechanics, and the payment obligations out of a dense agreement — maybe to build a closing checklist, abstract a lease, or compare against a prior version. By hand this is tedious and error-prone. AI pulls the structured details out quickly and consistently.

What that buys you. The mechanical extraction that used to take half an hour of careful scanning becomes a list you check rather than compile. For repeat document types, this is where the time savings are most consistent and least risky — the model is finding things that are explicitly in the text, not inferring or judging.

The boundary. Verify the list against the document, especially for anything you'll rely on. Extraction is accurate far more often than not, but "far more often than not" still means you confirm.

Capability 3: Summarizing and triaging long material

The workflow. A 60-page filing lands, or a client dumps a folder of correspondence on you, and you need to know what's in it before you can plan. AI produces a summary that lets you triage — what is this, what's the gist, where should I dig in.

What that buys you. Speed at the front of a matter, when you're deciding where to spend real attention. A good summary turns "I have to read all of this before I know what I'm dealing with" into "I know the shape of this, and now I'll read the parts that matter closely."

The boundary. A summary is a starting point for your reading, not a replacement for it. A fluent summary can quietly omit or soften the one clause that turns out to be decisive. Use it to prioritize, then read the source for anything you'll act on.

Capability 4: Answering questions about a matter

This is the capability that feels most like having an assistant who has read the file.

The workflow. Mid-matter, you have a specific question — "what's the notice period for termination?", "where does this agreement address assignment?", "did the other side ever put the indemnity cap in writing?" Instead of re-reading documents to find the answer, you ask, grounded in that matter's files, and get a pointed answer with a pointer to the source.

What that buys you. It collapses the re-reading tax. Over the life of a matter you return to the same documents repeatedly to answer small questions; being able to interrogate the file directly turns minutes of searching into seconds.

The boundary. Confirm the cited passage says what the answer claims. Grounded Q&A is far less prone to invention than open-ended generation, but you still check the source it points to — the value is that it points you to the right place fast, not that it relieves you of reading the place.

Capability 5: Drafting assistance — with sharp boundaries

Drafting is where AI's reputation and its real, narrow usefulness diverge most, so this one needs care.

What it's genuinely good for. Routine, non-confidential drafting: a plain-language first pass at a generic clause you'll adapt, a clearer version of a paragraph you wrote, turning your bullet points into prose, a template you'll fill in yourself, an explanation of a concept you can rework into client-facing language. For the shape and first-draft of non-sensitive text, it's a real time-saver.

The two boundaries that matter, both hard. First, confidentiality: drafting that involves a client's actual facts should not go through a cloud tool, because that transmits the material to a third party — keep client-specific drafting either out of cloud AI entirely or on a tool that doesn't transmit. Second, authorship and accuracy: you remain the author. AI cannot be trusted to produce reliable, citation-bearing legal documents on its own — it will invent authority that looks real — so a generated draft is raw material you verify and own, never a finished work product you file.

Used inside those boundaries, drafting assistance is useful. Pushed past them — feeding it client facts on a cloud tool, or filing what it produced without verifying — is exactly how the cautionary stories start.

Capability 6: Time entries and routine administrative text

The least glamorous capability is often the one that quietly recovers the most billable time.

The workflow. Solo practitioners are notorious for under-capturing time, because writing up entries is the chore you skip when you're busy. AI can turn your terse notes — "call w/ client re: lease, 25 min; reviewed redlines" — into clean, consistent narrative entries, and help keep your descriptions uniform across a matter.

What that buys you. Captured time you'd otherwise lose, and less end-of-month reconstruction. The work is generating readable text from your own notes, which is squarely the kind of bounded task AI handles well.

The boundary. It's drafting from your input, so the entry has to reflect what actually happened — you review for accuracy and for billing judgment, and anything touching a matter's specifics belongs on a tool that keeps the data on your machine rather than a public chatbot.

What AI can't — and shouldn't — do

Naming the limits isn't hedging; it's how you use the tool without getting burned. The honest list:

• It doesn't exercise legal judgment. It can flag that a clause is unusual; deciding what that means for your client, this deal, this jurisdiction is lawyering, and it stays with you. The model has no stake, no duty, and no understanding of consequences.
• It hallucinates. Language models produce fluent, confident text that can be flat wrong, including invented case citations. Every factual or legal claim is unverified until you confirm it against a primary source.
• It is not a legal-research tool. A general model has no reliable, maintained body of law behind it and will fabricate citations if asked for them. Document analysis is about your documents; legal research against real authority is a different job for a different tool, with your verification on top.
• It shouldn't run unsupervised. The value model is "assistant you check," not "associate you trust." Anything that leaves your office carries your name, not the model's.
• It can't be trusted with confidential material on cloud tools. A client's documents sent to a cloud service have been disclosed to a third party regardless of the provider's policies. Sensitivity of the material has to drive the choice of tool.
• It isn't a lawyer. It doesn't form a relationship with your client, hold a duty to them, or take responsibility for outcomes. It's software that reads fast.

None of this makes AI less useful. It makes the useful part legible: a tool that does the mechanical reading and writing quickly, under your supervision, on material you're allowed to give it.

What a realistic week looks like

None of these capabilities is dramatic on its own. The value shows up in aggregate, across an ordinary week.

A new engagement comes in with a stack of documents; you summarize them to triage before your intake call instead of reading blind. A client sends a vendor contract; you run a first-pass review, then spend your time on the two clauses that actually matter rather than all thirty pages. You're abstracting a lease, so you extract the key terms into a checklist instead of compiling it by hand. Mid-week, a question comes up on an older matter, and you ask the file directly instead of re-reading three documents to find the answer. Friday, you turn the week's scribbled notes into proper time entries before they evaporate.

No single step was transformative. But the reading-and-typing tax on the whole week dropped, and the judgment — which clause matters, what to advise, whether to sign — stayed exactly where it belongs. That's the realistic picture: not AI doing your job, but AI clearing the underbrush so you spend more of your day on the part only you can do.

Setting honest expectations

It's worth calibrating what "saves time" actually means, because oversold expectations are how people end up disappointed or, worse, careless.

AI is not going to halve your hours or run your practice. What it reliably does is compress specific mechanical tasks — the first-pass read, the manual extraction, the summarize-to-triage step, the write-up of routine text — from long to short. The judgment-heavy work that makes up the core of lawyering is roughly as long as it ever was, because that work is the value and AI can't do it for you.

There's also a real cost that the demos skip: verification takes time. A generated summary you still have to check, a draft you still have to own, an extraction you still have to confirm — the review is not free. The net win is real, but it's "the mechanical parts got much faster and the checking stayed," not "the work disappeared." Practitioners who account for the verification step up front stay both efficient and safe. Those who assume the output is finished are the ones who get surprised.

Held to that standard, AI is genuinely worth adopting for a solo practice — not as a miracle, but as a durable reduction in the drudgery that stands between you and the work only you can do.

How to fold this into a solo practice

You don't need to reorganize your practice to capture most of the value. A workable path:

1. Pick one document-heavy task you do often — first-pass contract review, or abstracting a recurring agreement type — and run AI on it for a few real matters.
2. Build the verification habit from day one. Treat every output as a draft to check against the source. The habit is what makes the speed safe.
3. Sort by sensitivity. Non-confidential drafting and general questions can go to ordinary tools; anything with a client's documents in it needs a tool that doesn't transmit them.
4. Keep it scoped to a matter. Working one matter at a time keeps it easy to reason about what the tool has seen and to check its work.
5. Expand only where it's earning its keep. Add tasks where the time savings are real and the verification is easy; don't force it onto work that needs judgment more than speed.

Getting good results: a few habits that help

The gap between "AI was useless" and "AI saved me an hour" is often just technique. A handful of habits reliably move you toward the second:

• Give it the actual document, not a description of it. These capabilities are strongest when the model is working from the real text in front of it rather than reasoning in the abstract — that's the difference between grounded help and guesswork.
• Ask for pointers, not verdicts. "Show me where termination is addressed and quote it" beats "is this a good contract." You want it to route you to the text; the assessment is yours.
• Be specific about the output you want. A checklist, a list of unusual clauses, a two-sentence summary per section — concrete asks get concrete, checkable answers.
• Iterate in follow-ups. Treat it like a conversation with an assistant: narrow, re-ask, drill into the section that matters. The second and third questions are usually where the value is.
• Always ask it to show its source in the document. Making it point to the passage both improves the answer and makes your verification faster.

None of this is special expertise. It's just the difference between pointing the tool at a bounded job and asking it to be a lawyer.

Where Privileged fits

Privileged is built for the analysis side of this list: document analysis and Q&A that runs entirely on-device via Ollama, organized by matter, with workflow templates for contract review, document summary, filing review, and time entry. It reads, reviews, extracts, summarizes, and answers questions about the documents you give it — all on your own machine, with nothing transmitted, retained off-device, or used to train anything.

What it is not, deliberately: it is not a legal-research tool, it doesn't look up case law, and it doesn't draft motions, briefs, or demand letters. Those are different jobs, and for anything you'll rely on as authority, the work is yours to verify. Privileged covers the bounded, confidential document work a solo practice runs on, where keeping the material on-device is the point. The how it works page has the specifics.

To go deeper on the capabilities here — how AI contract review actually works, concrete walkthroughs of reviewing an NDA and a commercial lease, and a fuller honest accounting of AI's limits in legal work — work through the guides in this cluster below.